The Andrysco et al. paper and algorithm is excellent. The only obvious improvement would be, in cases where there is more than one shortest unambiguous string,

The Andrysco et al. paper and algorithm is excellent. The only obvious improvement would be, in cases where there is more than one shortest unambiguous string, printing the one which is closest to the true floating-point value, which I don't believe the Errol algorithm does. But that's getting into serious hair-splitting territory.